Fields of \(u\)-invariant \(9\) (Q5960970)

Let \(F\) be a field of characteristic \(\neq 2\). One of the most important field invariants in the algebraic theory of quadratic forms is the \(u\)-invariant, which is defined to be the supremum of the dimensions of anisotropic finite-dimensional quadratic forms over \(F\), and the determination of the possible values of the \(u\)-invariant has been the focus of a large number of papers by various authors. It is not difficult to show that \(u\neq 3,5,7\), and it is easy to construct fields with \(u=2^n\) for any \(n\). \textit{I. Kaplansky} [J. Math. Soc. Japan 5, 200-207 (1953; Zbl 0051.02902)] conjectured that \(u\) (which he calls \(C\)) is always a power of \(2\). This conjecture was laid to rest when \textit{A. S. Merkur'ev} [Math. USSR, Izv. 38, 215-221 (1992); translation from Izv. Akad. Nauk SSSR, Ser. Mat. 55, 218-224 (1991; Zbl 0733.12008)] showed in a spectacular breakthrough that to every positive integer \(m\) there exists a field \(F\) with \(u(F)=2m\). His proof is based on his index reduction formulas for central simple algebras over the function field of a quadratic form, and the method of constructing the field became a model for many other constructions of fields having certain prescribed invariants pertaining to quadratic forms (for a survey, see the reviewer's article [Contemp. Math. 272, 73-102 (2000; Zbl 1025.11006)]). However, it remained open whether fields with odd \(u\)-invariant \(>1\) exist. The present paper by Oleg Izhboldin, who died tragically on April 17, 2000 at the age of 37 after submitting this article, constitutes major progress in the algebraic theory of quadratic forms. Not only does it prove the existence of fields with \(u\)-invariant \(9\), but the tools developed to achieve this goal also solve various problems in the theories of Chow groups and unramified cohomology of quadrics. As a bonus, the author proves several conjectures concerning the isotropy of quadratic forms over the function field of a quadric. The general idea of the construction of a field with \(u=n\) is to start with any field with a certain anisotropic form \(\phi\) of dimension \(n\), to take successively function fields of all forms of dimension \(n+1\) and then to pass to the limit. By construction, forms of dimension \(n+1\) over the field thus obtained will be isotropic. If one can show that the form \(\phi\) will stay anisotropic in the process, the field will have \(u\)-invariant \(n\). In the present paper, it therefore becomes necessary to study the (an)isotropy behaviour of \(9\)-dimensional forms over function fields of \(10\)-dimensional forms, and to find a suitable \(9\)-dimensional form which will stay anisotropic in the process. This role will be played by what the author calls an essential \(9\)-dimensional form. Such a form \(\phi\) is, by definition, of dimension \(9\) and anisotropic. It is not a so-called Pfister neighbor, i.e. it is not similar to an orthogonal summand of a \(4\)-fold Pfister form (a \(4\)-fold Pfister form is a tensor product of four binary forms), and the index of the even part of the Clifford algebra of \(\phi\) is \(\geq 4\). It is shown that essential \(9\)-dimensional forms stay essential (and hence anisotropic) over function fields of forms of dimension \(\geq 10\). Starting with such an essential form, the above construction will yield a field with \(u=9\). One of the key results used in the proof is due to \textit{N. A. Karpenko} [J. Pure Appl. Algebra 160, 195-227 (2001; Zbl 0998.11016)] stating that an essential \(9\)-dimensional form \(\phi\) over \(F\) becomes isotropic over \(F(\psi)\) with \(\dim\psi =9\) if and only if \(\phi\) is similar to \(\psi\). A major role in the proof is played by Chow groups and unramified cohomology of quadrics, and an important part is devoted to the computation of the torsion subgroup of the third Chow group of a quadric \(X_{\phi}\) for a form \(\phi\) of dimension \(\geq 9\). By work of Karpenko, it was known that \(\text{Tors CH}^3(X_{\phi}) = 0\) or \({\mathbb Z}/2{\mathbb Z}\), and that it is \(0\) if \(\dim\phi >12\). The author characterizes explicitly the cases where \(\text{Tors} CH^3\cong {\mathbb Z}/2{\mathbb Z}\) for \(9\leq\dim\phi\leq 12\), thus completing \textit{N. A. Karpenko}'s computation [Nova J. Algebra Geom. 3, 357-379 (1995; Zbl 0902.14006), Leningr. Math. J. 2, No. 1, 119-138 (1991); translation from Algebra Anal. 2, No. 1, 141-162 (1990; Zbl 0791.11019)] of CH\(^3(X_{\phi})\) for forms \(\phi\) except for the case of \(8\)-dimensional forms with nontrivial discriminant. The computation of \(\text{Tors CH}^3(X_{\phi})\) is of great interest in its own right as it is closely related to the fourth unramified cohomology group of quadrics. Let \(H^n(F)\) denote the \(n\)-th Galois cohomology group of \(F\) with \({\mathbb Z}/2{\mathbb Z}\)-coefficients, and let \(\widetilde{H}^4_{nr}(F(\phi)/F)\) denote the homology group of the complex \[ H^n(F)\to H^4(F(X_{\phi}))\to\bigoplus_{x\in X^{(1)}_{\phi}}H^3(F(x)) \] where \(F(x)\) is the residue field at the point \(x\) of codimension \(1\) in \(X_{\phi}\). \(\widetilde{H}^4_{nr}(F(\phi)/F)\) was studied by \textit{B. Kahn, M. Rost} and \textit{R. Sujatha} [Am. J. Math. 120, 841-891 (1998; Zbl 0913.11018)] and \textit{B. Kahn} and \textit{R. Sujatha} [Duke Math. J. 106, 449-484 (2001; Zbl 1049.11044)] where it was shown that it maps injectively into \(\text{Tors CH}^3(X_{\phi})\) for forms of dimension \(\geq 9\) which are not Pfister neighbors of \(4\)-fold Pfister forms, and that in this case it is an isomorphism provided that \(F\) contains all \(2\)-primary roots of unity. In the present paper, it is shown that the isomorphism also holds after one dispenses with the hypothesis on the roots of unity. In addition to their use in the construction of fields with \(u=9\), these methods from unramified cohomology and Chow groups of quadrics are then also applied to prove various conjectures about the isotropy behaviour of quadratic forms over function fields of quadrics, for example, that \(12\)-dimensional anisotropic forms in \(I^3\) (resp. \(10\)-dimensional anisotropic forms in \(I^2\) with \(\text{ind}(C(\phi))=2\)) stay anisotropic over function fields of forms of dimension \(>12\) (resp. \(>10\)). The author also characterizes \(10\)-dimensional anisotropic forms with maximal splitting. By a result of the reviewer [Math. Z. 220, 461-476 (1995; Zbl 0840.11017)], the Witt index of an anisotropic form of dimension \(2^n+m\) (\(1\leq m\leq 2^n\)) over its own function field is always \(\leq m\), and the form is said to have maximal splitting if equality holds. Anisotropic forms of dimension \(2^n+1\) and Pfister neighbors always have maximal splitting, and it is known that if \(2^{n+1}-m<\dim\phi\leq 2^{n+1}\) with \(m=\min\{ 2n , 8\}\), then maximal splitting implies that \(\phi\) is a Pfister neighbor by work of the author and \textit{A. Vishik} [Contemp. Math. 272, 103-125 (2000; Zbl 0972.11017)], so the first open case was that of dimension \(10\). In the present paper, it is proved that an anisotropic \(10\)-dimensional form has maximal splitting if and only if it is a Pfister neighbor or a product of a binary and a \(5\)-dimensional form. Finally, he characterizes nongood forms of height \(2\) and degree \(n=3\), i.e. anisotropic forms \(\phi\) over a field \(F\) whose anisotropic part over \(F(\phi)\) is similar to an \(n\)-fold Pfister form which cannot be defined over \(F\). It is shown that such a form is always a product of a binary form and a \(6\)-dimensional form in \(I^2\), thus confirming a conjecture by Kahn. The proof of this result relies on earlier work by Vishik which necessitates the field to be of characteristic \(0\). New methods developed by Vishik and by Karpenko after this paper was published imply that this result holds in any characteristic \(\neq 2\).